Mark Zuckerberg — Llama 3, $10B models, Caesar Augustus, & 1 GW datacenters

Mark Zuckerberg 0:00 That's not even a question for me, whether we're gonna go take a swing at building the next thing. I'm just incapable of not doing that. (whoosh) There's a bunch of times when we wanted to launch features and then Apple's just like, "Nope, you're not launching that." I was like, "That sucks."

Dwarkesh Patel 0:13 (laughs)

Mark Zuckerberg 0:14 Are we set up for that with AI where you're gonna get a handful of companies that run these closed models that are gonna be in control of the APIs, and therefore are gonna be able to tell you what you can build? (whoosh) Then when you start getting into building a data center that's like 300 megawatts or 500 megawatts or a gigawatt, just no one has built a single gigawatt data center yet. (whoosh) From wherever you sit, there's gonna be some actor who you don't trust. If they're the ones who have, like, the super strong AI, I think that that's potentially a much bigger risk.

Dwarkesh Patel 0:44 Mark, welcome to the podcast.

Mark Zuckerberg 0:45 Hey, thanks for having me. Big fan of your podcast.

Dwarkesh Patel 0:47 Oh, thank you. That's very nice of you to say. Um, (laughs) okay, so let's start by talking about the releases that will go out a l- when this interview goes out. Um, tell, tell me about the models, tell me about Meta AI. What's new? What's exciting about them?

Mark Zuckerberg 0:59 Yeah, sure. So, you know, I think the, the main thing that most people in the world are gonna see is the new version of Meta AI.

Dwarkesh Patel 1:04 Hmm.

Mark Zuckerberg 1:04 Right? So it's, um... And, you know, the most important thing about what we're doing is the upgrade to the model. We're rolling out Llama 3. We're doing it both as open source for the, the dev community, and it is now gonna be powering Meta AI. Um, so, you know, there's a lot that I'm sure we'll go into around Llama 3, but I think the bottom line on this is that with Llama 3, we now think that Meta AI is the most intelligent AI assistant that people can use that's freely available. Um, we're also integrating Google and Bing for real-time knowledge.

Dwarkesh Patel 1:34 Mm-hmm.

Mark Zuckerberg 1:34 Um, we're gonna make it a lot more prominent across our apps. So, you know, basically, you know, at the top of WhatsApp, and Instagram, and Facebook, and Messenger, uh, you'll just be able to, um, you know, use the search box right there to ask, ask us any question. Um, and there's a bunch of new creation features that we, that we added that I think are pretty cool that I think people enjoy. Uh, and I think animations is, is a good one. Um, you can basically just take any image and animate it. But I think one that, that, uh, people are gonna find pretty wild is, uh, it now generates high-quality images so quickly. I don't know if you've gotten a chance to play with this. That it actually generates it as you're typing and updates it in real time. So you're, like, typing your query and it's, and it's kind of, like, honing in on... And, and, you know, it's like, "Okay, here, um, you know, show me a picture of a, a cow, okay, in a field with mountains in the background." And just like everything's populated-

Dwarkesh Patel 2:25 Eating macadamia nuts. (laughs)

Mark Zuckerberg 2:26 Yeah, eating macadamia nuts, drinking beer, and like-

Dwarkesh Patel 2:27 (laughs)

Mark Zuckerberg 2:27 ... it just... And, and just like it's updating the image in real time. It's pretty wild. I think people are gonna enjoy that. Um, so yeah. So that, I think, is... That's what most people are gonna see in the world, right? We're rolling it out, um, you know, not everywhere, but we're starting in, um, in a handful of countries, and we'll do more over the coming weeks and months. Um, so that's, that, I think, is gonna be a pretty big deal. Um, and I'm really excited to get that in people's hands. It, it's, it's a big step forward for Meta AI. Um, but I think, you know, if you want to get under the hood a bit, the Llama 3 stuff is, is obviously the most technically interesting.

Dwarkesh Patel 3:01 Yeah.

Mark Zuckerberg 3:02 So, you know, we're, we're basically the... For the first version, we're training three versions. Um, you know, an eight billion and a 70 billion, which we're releasing today, and a 405 billion dense model, um, which is still training. So, so we're not releasing that today. Um, but, you know, the eight and 70, I mean, I'm, I'm pretty excited about how they turned out. I mean, it's, um, you know, they're, they're leading for, for their scale. Um, you know, it's, uh... I mean, we'll, we'll release a blog post with all of the benchmarks so people can check it out themselves. And obviously it's open source, so people get a chance to play with it. Um, we have a roadmap of new releases coming, uh, that are gonna bring multi-modality, more multilinguality, um, bigger context windows to those as well. Um, and then, you know, hopefully sometime later in the year, we'll, we'll get to roll out the 405, which I think is, is, um... You know, in training, it's still training, but, uh, for, for where it is right now in training, it is already at, um, around 85 MMLU-

Dwarkesh Patel 4:05 Mm-hmm.

Mark Zuckerberg 4:06 ... and, um, and just we, we expect that it's gonna have leading benchmarks on a, on a bunch of, on a bunch of the, the benchmarks. So I, I'm, I'm pretty excited about all of that. I mean, the 70 billion is, is, um, is, is great too. I mean, we're releasing that today. It's around 82 MMLU and has leading scores on math and reasoning. So, I mean, it's, uh... I think just getting this in people's hands is gonna be pretty wild.

Dwarkesh Patel 4:26 Oh, interesting. Yeah, that's-

Mark Zuckerberg 4:27 Yeah.

Dwarkesh Patel 4:27 ... the first time hearing this benchmarking. That's super impressive. And it'll be-

Mark Zuckerberg 4:29 Yeah, yeah, the eight billion is, the eight billion is, um, is nearly as, as powerful as the biggest version of Llama 2 that we released.

Dwarkesh Patel 4:38 Hmm.

Mark Zuckerberg 4:38 So it's like the smallest Llama 3 is basically as powerful as the, the biggest Llama 2.

Dwarkesh Patel 4:42 Hmm, okay, so before we dig into these models-

Mark Zuckerberg 4:46 Yeah.

Dwarkesh Patel 4:46 ... I actually wanna go back in time. 2022 is, I'm assuming, when you started acquiring these H100s, um, or but you, you can tell me when.

Mark Zuckerberg 4:54 Yeah, yeah.

Dwarkesh Patel 4:54 Um, where you're like, "Stock price is getting hammered." People are like, "What's happening with all this CapEx?" People aren't buying the metaverse, and presumably you're spending that CapEx to get these H100s. How... Back then, how did you know to get the H100s? How did you know it- you'll need the GPUs?

Mark Zuckerberg 5:06 Um, I think it was, it was because we were working on Reels. So, you know, we got into this situation where, um, you know, we always wanna have enough capacity to build something that we can't quite see that were on the horizon yet. Um, and we got into this position with Reels where we needed more GPUs to train the models, right? It was, it was this big evolution for services where instead of just ranking content from people who you follow or your friends and, and whatever pages you follow, um, we made this big push to basically start recommending what, what we call unconnected content. Basically connect, uh, content from people or pages that you're not following. So now, kind of the, the corpus of, of kind of content candidates that we could potentially show you expanded from, you know, on the order of thousands to on the order of hundreds of millions. So completely different infrastructure.... and we, um, started working on, on doing that and we were constrained on, um, on basically the infrastructure that we had to catch up to what TikTok was doing as quickly as we would've wanted to. Um, so I basically looked at that and I was like, "Hey, we have to make sure that we're never in this situation again, so let's order enough GPUs to do what we need to do on Reels and ranking content in feed, but let's also... let's double that." Right? 'Cause a- again, like our normal principle is there's gonna be something on the horizon that we can't see yet.

Dwarkesh Patel 6:35 Did you know it would be AI?

Mark Zuckerberg 6:36 Um, well, we thought it would be... we thought it was gonna be something that had to do with training large models, right? I mean, but at the time, I, I thought it was probably gonna be more something that had to do with content. But I don't know. I mean, it's, it's almost just the pattern matching in running the company is there's always another thing. Right? So it's... I, I, I'm not even sure I had... at that time, I was so deep in just, you know, trying to get, you know, the recommendations working for Reels and, and other content 'cause I mean, that's just such a big unlock for Instagram and Facebook, now being able to show people content that's interesting to them that they're... from people that they're not even following. But, um, yeah. I- I... that, that ended up being a very good decision in retrospect.

Dwarkesh Patel 7:16 Yeah. Yeah.

Mark Zuckerberg 7:16 Yeah.

Dwarkesh Patel 7:17 Okay. And then-

Mark Zuckerberg 7:17 But it came from being behind. Not-

Dwarkesh Patel 7:18 So then-

Mark Zuckerberg 7:18 It wasn't like I was... I... you know, it wasn't like, oh, I was so far ahead. Actually, most of the times, I think, where we kinda make some decision that ends up seeming good is because we messed something up before and just didn't wanna repeat the mistake.

Dwarkesh Patel 7:31 Uh, this is a total detour-

Mark Zuckerberg 7:32 Yeah.

Dwarkesh Patel 7:32 ... but I actually want to ask about this while we're on this. We'll get back to ANI in a... AI in a second. So, you didn't sell for one billion, but presumably there's some amount you would've sold for, right? Did you write down in your head, like, "I think the actual valuation of Facebook at the time is this," and they're not actually getting the valuation right? I mean, they're already $5 trillion. Of course you would have sold. So, wh- wh- like, wh- how did you think about that choice?

Mark Zuckerberg 7:52 Yeah, I don't know. I mean, look, I think some of these things are just personal. Um, I, I don't know at the time that I was sophisticated enough to do that analysis.

Dwarkesh Patel 8:02 Mm.

Mark Zuckerberg 8:02 But I had all these people around me who were making all these arguments for how, like, a billion dollars was... you know, it's like here's the revenue that we need to make and here's how big we need to be and, like, it's clearly so many years in the future. Like, and it was, it was very far ahead of where we were at the time. And I don't know, I didn't, I didn't really have the financial sophistication to really even engage with that kind of debate. I just... I, I think I sort of deep down believed in what we were doing, and I, I did some analysis. Um, I was like, "Okay, well, what would I go do if I wasn't doing this?" It's like, well, I really like building things and I like helping people communicate and I like understanding what's going on with people and the dynamics between people, so I think if I sold this company I'd just go build another company like this, and I kinda like the one I have.

Dwarkesh Patel 8:54 (laughs)

Mark Zuckerberg 8:54 So, um, so I mean, you know, what's... wh- wh- why, right? But, um, I don't know. I, I, I think a lot of the biggest bets that people make, um, are often just based on conviction and values. Um, not n- it's, it's actually usually very hard to do the analyses trying to connect the dots forward.

Dwarkesh Patel 9:14 Yeah.

Dwarkesh Patel 9:15 So, you've had, um, Facebook AI Research for a long time. Uh, now it's become seemingly central to your company. At what point did making AGI or whatever, however you consider that mission, at what point is that like, this is a key priority of what Meta is doing?

Mark Zuckerberg 9:33 Yeah. I mean, it's been a big deal for a while.

Dwarkesh Patel 9:35 Right.

Mark Zuckerberg 9:35 So we, we started FAIR, um, about 10 years ago, and the idea was that along the way to general intelligence or AI, like full AI, whatever you wanna call it, there are gonna be all these different innovations and that's gonna just improve everything that we do. So, it... I... we didn't kind of conceive it as a product. It was more kind of a research group.

Dwarkesh Patel 9:57 Yeah.

Mark Zuckerberg 9:58 And over the last 10 years, it has created a lot of different things that have basically improved all of our products, um, and advanced the field and allowed other people in the field to create things that have improved our products too. So, I think that that's been great. But there's obviously a big change, um-

Dwarkesh Patel 10:16 Yeah.

Mark Zuckerberg 10:17 ... in the, in the last few years when, you know, ChatGPT comes out, um, the diffusion models around image creation come out and, like, I mean, this is some pretty wild stuff, right? That, that I think is, like, pretty clearly gonna affect how, how people interact with, like, every app that's out there. So I... at that point, we started a second group, um, the, the GenAI group, um, with the goal of basically bringing that stuff into our products, so building leading foundation models that would, that would sort of power all these different products. And initially when we started doing that, um, the theory at first was, hey, a lot of this stuff that we're doing is, is pretty social, right? So, you know, it's helping people interact with creators, helping, um, people interact with businesses to... you know, so that businesses can sell things or do customer support or, um, you know, basic assistant functionality for, um... you know, whether it's for our apps or the smart glasses or, or VR, like all these different things. So initially it wasn't completely clear that you were gonna need kind of full AGI, um, to be able to support those use cases. But then through working on them, I think it's actually become clear that you do, right? There... in all these subtle ways. So for example, you know, for LLaMA 2, when we were working on it, we didn't prioritize coding. And the reason why we didn't prioritize coding is because people aren't gonna ask Meta AI a lot of coding questions in WhatsApp, right?

Dwarkesh Patel 11:43 Now they will. (laughs)

Mark Zuckerberg 11:44 Well, I don't know. I'm, I'm not sure that WhatsApp is like the UI that people are gonna be doing a lot of coding questions.

Dwarkesh Patel 11:49 Yeah.

Mark Zuckerberg 11:49 So we're like, all right, look, in terms of the things that... you know, or, or Facebook or Instagram or, you know, those, those different services. Maybe, maybe the website, right? Meta dot a- dot AI that we're, we're launching, I think.

Dwarkesh Patel 11:57 Yeah.

Mark Zuckerberg 11:58 But, but the, the thing that was sort of, uh, uh, I think has, has been a, you know, somewhat surprising result over the last, um, you know, 18 months is that...... it, it turns out that coding is important for a lot of domains, not just coding, right? So even if people aren't asking coding questions to the models, um, training the models on coding helps them, um, just be more rigorous and answer the question and, and kind of, um, help reason across a lot of different types of domains. Okay, so that's one example where it's like, all right, so for Llama 3, we, like, really focused on training it with a lot of coding because it's like, all right, that's gonna make it better on all these things even if people aren't answering... aren't asking primarily coding questions. Reasoning I think is another example. It's like, okay, yeah, maybe you wanna chat with a creator or, you know, you're a business and you're trying to interact with a customer. You know, that interaction is not just like, okay, the person sends you a message and you just reply, right? It's a, it's like a multi-step interaction where you're trying to think through, "How do I accomplish the person's goals?" And, um, you know, a lot of times when a customer comes, they don't necessarily know exactly what they're looking for or how to ask their questions, so it's not really the job of the AI to just respond to the question. It's like you need to kinda think about it more holistically. It's really... It becomes a reasoning problem, right? So if someone else, you know, solves reasoning or makes good advances on reasoning and we're sitting here with a basic chatbot, then like, our product is lame compared to what other people are building. So it's like, so, okay, so at the end of the day, we've got, we, you know, I, we basically realized we've gotta solve general intelligence, um, and we just kinda upped the ante and the investment to make sure that we could do that.

Dwarkesh Patel 13:32 So the version of Llama that, um, that, uh, that's going to solve all these use cases for users, is that the version that will be powerful enough to, like, replace a programmer you might have in this building?

Mark Zuckerberg 13:46 I mean, I just think that all this stuff is gonna be progressive over time.

Dwarkesh Patel 13:49 But, uh, end case, Llama 10.

Mark Zuckerberg 13:51 Um, I, I mean, I think that there's a lot baked into that question. I, I'm not sure that we're replacing people as much as-

Dwarkesh Patel 13:59 M- maybe-

Mark Zuckerberg 14:00 ... giving people tools to do more stuff, but, um-

Dwarkesh Patel 14:02 Is the programmer in this building 10X more productive after Llama 10? (laughs)

Mark Zuckerberg 14:04 I would hope more, but, um, but no. I mean, look, uh, I, I'm not... I don't believe that there's, like, a single threshold of intelligence for, for humanity because, I mean, people have different skills. And at some point, I think that AI is gonna be, um... is, is probably gonna surpass people at most of o- of those things, um, depending on how powerful the models are. But, um, but I think it's progressive, and I, I don't think AGI is one thing. I think it's... you're basically adding different capabilities. So multimodality is, is kind of a key one that we're focused on now, initially with photos and images and text but eventually with videos. And then because we're so focused on the metaverse, kind of 3D type stuff is important. Um, one modality that I'm pretty focused on that I haven't seen as many other people in the industry, um, focus on this is sort of, like, emotional understanding. Like, I mean, uh, so much of, of the human brain is just dedicated to understanding people and, and kind of like understanding your expressions and emotions, and I think that that's like its own whole modality, right? That, um... I mean, you could say, okay, maybe it's just video or image, but it's, like, clearly a very specialized version of those two. So there's all these different capabilities that I think you wanna basically train the models to focus on as well as getting a lot better at reasoning, getting a lot better at memory, which I think is, is kind of its own whole thing. It's... I mean, I don't think we're gonna be, you know, primarily shoving context or, or, or kind of things into a query context window, um, in the future to, to ask more complicated questions. I think that there will be kinda different stores of memory or different custom models that, um, that are maybe more personalized to people. But I don't know. I think that these are all just different capabilities. And then obviously making them big and small. We care about both because, you know, we wanna... You know, if you're running something like Meta AI, then we have the ability to... That's pretty server-based, um, but we also want it running on smart glasses, and, you know, there's not a lot of space in smart glasses so, um, you, you wanna have something that's very efficient for that.

Dwarkesh Patel 16:00 What is the use case that if you're doing tens of billions of dollars' worth of inference or even eventually hundreds of billions of dollars' worth of inference, if you're using intelligence in an industrial scale, what is a use case? Is it simulations? Is it the AIs that will be in the metaverse? What, what are we using the data centers for?

Mark Zuckerberg 16:16 Um... I mean, our bet is that it's gonna... this is basically gonna change all of the products, right? So I, I think that there's gonna be a kind of meta AI general assistant product, and I think that that will shift from something that feels more like a chatbot where it's like you just ask a question and it kind of formulates an answer to things where you're increasingly giving it more complicated tasks and then it goes away and does them.

Dwarkesh Patel 16:41 Mm.

Mark Zuckerberg 16:42 So that's gonna take a lot of inference. It's gonna take a lot of compute in other ways, too. Um, then I think that there's a big part of what we're gonna do that is, um, like interacting with other agents for other people, so whether it's businesses or creators. Um, I guess a big part of my theory on this is that there's not just gonna be, like, one singular AI that you interact with because I think, um, you know, every business is gonna, like, want an AI that represents their interests. They're not gonna, like, wanna primarily interact with you through an AI that is gonna sell their competitors' customers. So... Uh, sorry, their competitors' products. Um, so, um, uh... So yeah, so I think creators is gonna be a big one and we- there are about 200 million creators on our platforms. They all basically have the pattern where, um, they want to engage their community but they're limited by hours in the day, and their community generally wants to engage them but they don't have... they're limited by hours in the day. Um, so if you could create something where, um, an AI could basically work... that creator can basically own the AI and train it in the way that they want, um, and can engage their community, I, I think that that's gonna be super powerful too. So, um, so I think that there's gonna be a ton of engagement across all these things. Um, but these are just the consumer use cases. I mean, I think when you think about stuff like... I mean, when I run, like, our foundation, right-... Chan Zuckerberg Initiative with my wife and, you know, we're doing a bunch of stuff on science and, um, and there's obviously a lot of AI work that, where I th- that I think is gonna advance science and healthcare and all these things too. So, I think that it's like, there's a, this is, I think gonna end up affecting basically every area of the products-

Dwarkesh Patel 18:23 Yeah.

Mark Zuckerberg 18:23 ... and, and, and the, and the, uh, the economy.

Dwarkesh Patel 18:25 The thing you mentioned about an AI that can just go out and do something for you that's multi-stuff, is that a bigger model? Is that, you'll make, like LLaMA 4 will still, uh, there'll be a version that's still 70B but will just be, you'll just train it on the right data and that will be super powerful? How do, like, what does the progression look like? Is it scaling? Is it just same size but different banks, like you were talking about?

Mark Zuckerberg 18:46 Um, I, I don't know that we know the answer to that. So I think one thing that is, seems to be a pattern is that you have the LLaMA, uh, sorry, the, the, the LLaMA model, and then you build some kind of other application-specific code around it, right? So some of it is, is the fine-tuning for the use case, but some of it is just, like, logic for, okay, how, um, like, how meta I should integrate, like should work with tools like Google or Bing to bring in real-time knowledge. I mean, that's not part of the base LLaMA model. That's like part of a... Okay, so for LLaMA 2, we had some of that, and it was a little more kind of hand engineered. And then part of our goal for LLaMA 3 was to bring more of that into the model itself. And, but for LLaMA 3 is we start getting into more of these agent-like behaviors.

Dwarkesh Patel 19:39 Mm-hmm.

Mark Zuckerberg 19:40 I think some of that is gonna be more hand engineered, and then I think our goal for LLaMA 4 will be to bring more of that into the model. So I think at each point, uh, of like, at each step along the way, you kind of have a sense of what's gonna be possible on the horizon. You start messing with it and hacking around it, um, and then I think that that helps you hone your intuition for what you wanna try to cha- train into the next version of the model itself.

Dwarkesh Patel 20:03 Interesting.

Mark Zuckerberg 20:04 Which makes it more general because obviously anything that you're hand coding is, um, you know, you can unlock some use cases, but it's just inherently brittle and non-general.

Dwarkesh Patel 20:12 Hmm. Hey, everybody. Real quick, I wanna tell you about a tool that I wish more applications used. So obviously you've noticed every single company is trying to add an AI chatbot to their website. But as a user, I usually find them really annoying 'cause they give these long, generic, often useless answers. Commandbar is a user assistant that you can just embed into your website or application, and it feels like you're talking to a friendly human support agent who is browsing with you and for you. And it's much more personalized than a regular chatbot. It can actually look up users' history and respond differently based on that. It can use APIs to perform actions. It can even proactively nudge users to explore new features. One thing that I think is really cool is that instead of just outputting text, Commandbar can kinda just say, "Here, let me show you," and start browsing alongside the user. Anyways, they're in a bunch of great products already. You can learn more about them at commandbar.com. Thanks to them for sponsoring this episode, and now back to Mark. When you say into the model itself, you train it on the thing that you want in the model itself? What do you mean by in, into the model itself?

Mark Zuckerberg 21:26 Well, I mean, I think like the, the example that I gave for LLaMA 2 where, you know, it's, we, we really, I mean for LLaMA 2 the tool use was very, very specific.

Dwarkesh Patel 21:38 Yeah.

Mark Zuckerberg 21:38 Um, whereas LLaMA 3 has the ability to, has much better tool use, right? So, so we don't have to like hand code all the stuff to have it use Google to, to go do a search. Um, it just kinda can do that. Um, so, and similarly for coding and, and kind of running code and just a, a bunch of stuff like that and, um, but I think once you kinda get that capability, then you get a peak of, "Okay, well what can we start doing next? Okay, well I don't necessarily wanna wait until LLaMA 4 is around to start building those capabilities, so let's start hacking around it." And-

Dwarkesh Patel 22:10 Mm-hmm.

Mark Zuckerberg 22:11 Um, so you do, you do s- a bunch of hand coding and that makes the, um, the products better if for the interim, but then that also helps show the way of what we wanna try to build into the next version of the model.

Dwarkesh Patel 22:21 What is the community fine tune of LLaMA 3 you're most excited by? Maybe not the one that will be most useful to you, but just you will just enjoy playing it with the most. They like fine tune it on antiquity and you'll just be like talking to Virgil or something. What, what are you excited about?

Mark Zuckerberg 22:34 I don't know. Um, I mean, I think the nature of the stuff is it's like, you get surprised, right? So I think like any, any specific thing that I sort of thought would be valuable, we'd probably be building.

Dwarkesh Patel 22:49 (laughs)

Mark Zuckerberg 22:49 Right? So, um, but I, I think you'll get distilled versions. I think you'll get kind of smaller versions. I mean, I mean one thing that, that I think is, eight billion I don't think is quite small enough for, for a bunch of use cases, right?

Dwarkesh Patel 23:05 Mm-hmm.

Mark Zuckerberg 23:05 I think like over time I'd love to get, you know, a billion parameter model or a two billion parameter model or even like a, I don't know, maybe like a 500 million parameter model and see what you can do with that because I mean, as, as they start getting, if, if it, if with eight billion parameters we're basically nearly as powerful as the largest LLaMA 2 model, then with a billion parameters you should be able to do something that's interesting, right, and faster, um, good for classification or a lot of kind of like basic things that people do before, um, kind of understanding the intent of a, of a user query and feeding it to the most powerful model to, to kind of hone what, what the, what the prompt should be. Um, so I don't know. I think that that's one thing that maybe the community can help fill in, but I mean, uh, we'll, we'll also, we're also thinking about getting around to distilling some of these ourselves, but, uh, right now the GPUs are, uh-

Dwarkesh Patel 23:54 (laughs)

Mark Zuckerberg 23:54 ... pegged training the 405, so.

Dwarkesh Patel 23:56 What, okay, so you have all these GPUs. Um, uh, th- uh, I think you said 350,000 by the end of the year.

Mark Zuckerberg 24:02 That's the whole fleet. I mean, I was-

Dwarkesh Patel 24:04 Yeah, yeah.

Mark Zuckerberg 24:04 ... uh, we, we built...... two, I think it's like 22, 24,000 clusters that are kind of the single clusters that we have for training the big models.

Dwarkesh Patel 24:14 Yeah.

Mark Zuckerberg 24:14 I mean, I mean, obviously, across a lot of the stuff that we do. A lot of our stuff goes towards training like Reels models, and-

Dwarkesh Patel 24:19 Sure.

Mark Zuckerberg 24:19 ... and like Facebook news feed and Instagram feed, and then inference is a huge thing for us 'cause we serve a ton of people, right? So, our ratio of inference compute required to, um, to training is probably much higher than most other companies that are doing this stuff just because the sheer volume of the community that we're serving, so.

Dwarkesh Patel 24:39 Yeah.

Mark Zuckerberg 24:39 Yeah.

Dwarkesh Patel 24:39 Yeah, that was really interesting in the material they shared with me before, that you trained it on more data than is compute optimal just for training because the inference is such a big deal for you guys and also for the community, that it makes sense to just have this thing to have trillions of tokens in there.

Mark Zuckerberg 24:52 Yeah, yeah. Although... And one of the interesting things about it, so we saw even with the 70 billion, is we, we thought it would get more saturated at, um, at... You know, it's like we trained it on around 15 trillion tokens, and-

Dwarkesh Patel 25:04 Yeah.

Mark Zuckerberg 25:05 ... we, I guess our prediction going in was that it was gonna asymptote more, but even by the end, it was still learning, right? It's like e- we, we probably could have fed it more tokens and it would have gotten somewhat better, but I mean, at some point, you know, you're running a company, you need to do these meta reasoning questions of like, "All right, how, do I wanna spend our GPUs on like training this 70 billion model further? Do we wanna kind of get on with it so we can start testing hypotheses for LLaMA-4?" So, we kinda needed to, to make, um, to make that call and I think we got it. We... I think we got to a reasonable balance for, for this version of the 70 billion. Um, there'll be others in the future where, you know, the 70 billion multimodal one that'll come, um, over the next period. But, um, but yeah, I mean, it's, it... That was, that was fascinating that you could just... That, that it's the architectures at this point can just take so much data.

Dwarkesh Patel 25:55 Yeah, that's really interesting. So, what does this imply about future models? Uh, you mentioned that the LLaMA-3-8B is better than the LLaMA-2-70B, right?

Mark Zuckerberg 26:03 No, no, no. It's nearly as good.

Dwarkesh Patel 26:04 Okay. (laughs)

Mark Zuckerberg 26:05 (laughs) It's... I don't, I don't overstate.

Dwarkesh Patel 26:06 But does that mean like

Dwarkesh Patel 26:07 the LLaMA-4-

Mark Zuckerberg 26:07 It's, it's similar to our magnitude.

Dwarkesh Patel 26:08 Does that, does that mean the LLaMA-4-70B will be as good as the LLaMA-3-405B? Like, or, uh-

Mark Zuckerberg 26:12 I mean, this is one of the-

Dwarkesh Patel 26:13 What is the future of this look like?

Mark Zuckerberg 26:13 This is one of the great questions, right?

Dwarkesh Patel 26:15 Yeah.

Mark Zuckerberg 26:15 That I think no one knows. Um, is, is basically, uh, you know, it's, it's one of the trickiest things in the world to plan around is when you have an exponential curve, how long does it keep going for?

Dwarkesh Patel 26:27 Yeah.

Mark Zuckerberg 26:27 And, um, I think it's likely enough that it will keep going, that it is worth investing the, um, tens or, you know, 100 billion plus in building the infrastructure to, um, assume that if that kinda keeps going, you're gonna get some really amazing, uh, things that are just gonna make amazing products.

Dwarkesh Patel 26:46 Mm.

Mark Zuckerberg 26:47 But I don't think anyone in the industry can really tell you that it will continue scaling at that rate for sure. Right? In general, you know, in history, you hit bottlenecks at certain points and now there's so much energy on this that maybe those bottlenecks get knocked over pretty quickly. But, um, but I don't know. I think that that's, that's an interesting question.

Dwarkesh Patel 27:08 What does the world look like where there aren't these bottlenecks? Let's, you know, suppose like progress just continues, uh, at, at this pace, which seems like plausible. Um, like zooming out-

Mark Zuckerberg 27:17 Well, they're gonna be di-

Dwarkesh Patel 27:18 ... but forgetting about like LLaMAs.

Mark Zuckerberg 27:18 They're gonna be different bottlenecks.

Dwarkesh Patel 27:20 Right. So if not training, then infor- like, oh, yeah. Go ahead.

Mark Zuckerberg 27:23 Well, I think at some point, you know, over the last few years, I think there's this issue of, um, GPU production.

Dwarkesh Patel 27:31 Yeah.

Mark Zuckerberg 27:31 Right? So, even companies that had the models, uh, so- sorry, that had the money to pay for the GPUs, um, couldn't necessarily get as many as they wanted because there was, there were all these supply constraints.

Dwarkesh Patel 27:42 Yeah.

Mark Zuckerberg 27:43 Now, I think that's sort of getting less. So now, I think you're seeing a bunch of companies think about, "Wow, we should just like really invest a lot of money in building out these things." And I think that that will go for, um, for some period of time. Um, I think there's a, there is a capital question of like, okay, at what point does it stop being worth it to put the capital in? But I actually think before we hit that, you're gonna run into energy constraints, right? Because, um, I just, I mean, I don't think anyone's built a gigawatt single training cluster yet, right? And, um, and then you run into these things that just end up being slower in the world, like getting energy permitted is like a very heavily regulated government function, right? So you're going from, on the one hand, software, which is somewhat regulated. I, I'd, I'd argue that it is more regulated than I think a lot of people in the, in the, in the tech community feel. Although it's obviously different. If you're starting a small company, maybe you feel that less. If you're a big company, you know, we just interact with people, but different governments and regulators are, you know, we have kinda lots of rules-

Dwarkesh Patel 28:52 Yeah.

Mark Zuckerberg 28:52 ... so, or that we need to kind of follow and make sure we do a good job with around the world. Um, but I think that there's no doubt that like energy, and if you're talking about building large new power plants or large build outs and then building transmission lines that cross other private or public land, that is just a heavily regulated thing. So you're talking about many years of lead time. So if we wanted to stand up just some like massive facility, um, to power that, I, I think that that is... that's, that's a very long term project. Right? And, um, so I don't know. I, I think that that's... I, I think people will do it. I don't... But, but I don't think that this is like something that can be quite as magical as just like, okay, you get a level of AI and you get a bunch of capital and you put it in and then like all of a sudden the models are just gonna kinda like just... Like I think you, you do hit different bottlenecks along the way.

Dwarkesh Patel 29:43 Yeah. Is there something, a project, maybe AI-related and maybe not, that even a company like Meta doesn't have the resources for? Like if your R&D budget or capex budget was 10X what it is now, then you could pursue it, like it's in the back of your mind. But M- Meta today, and m- maybe you could like... 'cause even if you can't even issue stock or bond for it, it's like just 10X bigger than your budget.

Mark Zuckerberg 30:03 Well, I think energy is one piece.

Dwarkesh Patel 30:04 Yeah.

Mark Zuckerberg 30:05 Right? Um, I think we would probably build out bigger clusters than we currently can.

Narrator 30:13 ... if we could get the energy to do it. So, I think that that's, um, eh, that- that's fundamentally f- uh, money bottlenecked in the limit? Like, if you had a trillion dollars-

Mark Zuckerberg 30:23 I think it's, I think it's time.

Narrator 30:24 Yeah.

Mark Zuckerberg 30:24 Right? Um, well if you look at it in terms of ... But it depends on how far the, the exponential curves go, right? Like, I think a number of companies are working on ... You know, right now I think of, you know, like, a lot of data centers are on the order of 50 megawatts or 100 megawatts, or like a big one might be 150 megawatts. Okay, so you take a whole data center and you fill it up with just all the stuff that you need to do for training and you build the biggest cluster you can. I think you're- y- you, that's kind of ... I think a bunch of companies are running at stuff like that. Um, but then when you start getting into building a data center that's like 300 megawatts or 500 megawatts or a gigawatt, uh, just, I mean, just no one has built a single gigawatt data center yet. So, I think it will happen, right? I mean, this is only a matter of time, but it's, it's not gonna be like next year, right? It's, um ... I think that some of these things will take, I don't know, some, some number of years to, to build out. And then the question is, okay, well, if you ... I mean just to I guess put this in perspective, I think a gigawatt, it's like around the size of like a meaningful nuclear power plant only going towards training a model. Didn't e- didn't Amazon do this? There's like a, they have a 950 giga- uh, megawatt, uh, thing. Yeah, I'm not, I'm not exactly sure what you did. You'd have to, wh- what they did, you'd have to ask them, um, um- But it doesn't have to be in the same place, right? If distributed training works, it can be distributed- Well, that, that I think is a big question. Yeah. Right? Is, is basically how that's gonna work. And I, I do think in the future, it seems quite possible that more of what we call training for these big models is actually more along the lines of inference generating synthetic data to then go feed into the model. So, I don't know what that ratio is gonna be, but I, I consider, um, the generation of synthetic data to be more inference than training today. But obviously if you're doing it in order to train a model, it's, it's part of the broader training process. So, um, I don't know. That's an o- that's a, an open question, is to, to kind of where, what the balance of that and how that plays out. If that's the case, would that potentially also be the case with Llama 3, and maybe like Llama 4 onwards, where you put this out and if somebody has a ton of compute, then using the models that you've put out you can just keep making these things arbitrarily smarter? Like s- uh, some, uh, uh, K- Kuwait or UAE or some random country has a ton of compute, um, and th- they can just, uh, actually just use Llama 4 to just make something much smarter. Um, I, I do think that there are gonna be dynamics like that. But I, I also think that there is a fundamental limitation on, um, on kind of the network architecture, right? Or the, the kind of model architecture, right? So, I think like a 70 billion model that kind of we trained with the Llama 3 architecture can get better, right? It can, it can keep going. Like I, like I was saying, it's- Yeah. ... you know, we felt like if we kept on feeding it more data or, or rotated the high value tokens through again then, then, you know, it would, it would continue getting better. But ... And- and we've seen a bunch of other people around the world, um, you know, different companies basically take the Llama 2 70 billion base, like take that model architecture and then build a new model. Um, it's still the case that when you make a generational improvement to the kinda Llama 3 70 billion or the Llama 3 405, there's nothing open source anything like that today, right? Like a, it- it's, it's not ... I think that that's like, it's a big step function in what people are gonna be able to build on top of that I don't think can go infinitely from there. I think it can, there can be some optimization in that until you get

Mark Zuckerberg 34:03 to the next step function. Yeah. Okay. L- so let's zoom out a little bit from, uh, specific models and even the many years lead times you would need to get energy approvals and so on. Like big picture. Yeah. These next couple of decades. Sure. What's happening with AI? Um, is, does it feel like another technology like metaverse or social or does it feel like a fundamentally different thing in the course of human history? Um, I think it's gonna be pretty fundamental. I think it's gonna be more like the creation of computing in the first place, right? So, um, you'll get all these new apps in the same way that when you got the web or you got mobile phones, you got ... Like people basically rethought all these experiences and a lot of things that weren't possible before now became possible. Um, so I think that will happen, but I think it's a much lower level innovation. It's, um, it's, it's gonna be more like going from people didn't have computers to people have computers is my, my sense. Um, but it's also, it's, it's, uh ... I don't know. It's, it's very hard to reason about exactly how this goes. I, I tend to think that, you know, in like the cosmic scale obviously it'll happen quickly over a, you know, couple of decades or something. But I, I do think that there, there is some set of people who are afraid of like, you know, really just kind of spins and goes from being like somewhat intelligent to extremely intelligent overnight, and I just think that there's all these physical constraints that make that so that that's unlikely to happen. Um, I, I, I just don't, I don't really see that, that playing out. So I think you'll have, I think we'll have time to kind of acclimate a bit, but it will really change the way that we work and give people all these creative tools to do different things that they ... Yeah. I, I think, I think it's gonna be, it's- it's gonna really enable people to do the things that they want a lot more is- Mm-hmm. ... is, is my view. Um, eh, uh, okay. So maybe not overnight, but i- is it your view that, like, on a cosmic scale if you think like humans evolved- Uh-huh. ... and then like AI happened and then they like went out through the galaxy? Or maybe it takes many decades. Maybe it takes a century, but like billi- is that like the grand scheme of what's happening right now in history?Um, uh, w- sorry, in what sense? I mean-

Dwarkesh Patel 36:24 In the sense that there are other technologies like computers and even, like, fire, but, like, w- i- the AI happening is as significant as, like, humans evolving in the first place.

Mark Zuckerberg 36:32 I, I think that that's tricky. Um, I think people like to, you know, I mean, the history of humanity, I think, has been people basically, you know, thinking that certain aspects of humanity are, like, really unique in different ways and then coming to grips with the fact that that's not true, but humanity is actually still super special. Right? So it's, um, you know, it's like we thought that the earth was the center of the universe, and it's like, it's not, but, like, it's, like, humans are still pretty awesome-

Dwarkesh Patel 37:09 Yeah.

Mark Zuckerberg 37:09 ... right? And, and pretty unique. Um, I think that another bias that people tend to have is thinking that intelligence is somehow kind of fundamentally connected to life, and it's not actually clear that it is, right? I, I think, like, like, people think that, um, I mean, I, I don't know that we have a clear enough definition of consciousness or, um, or, or, or life to kinda fully, um, interrogate this. But I know there were, there's all this science fiction about, okay, you create intelligence, and now it, like, starts taking on all these human-like behaviors and, and things like that, but I actually think that the current incarnation of all this stuff at least kind of feels like it's going in a direction where intelligence can be pretty separated from consciousness and agency and things like that, that, um, I think just makes it a super

Mark Zuckerberg 38:04 valuable tool. So I, I don't know. I mean, obviously, it's, it's, um, it's very difficult to predict what direction this stuff goes in over time, which is why I, I don't think anyone should be dogmatic about, you know, how they plan to develop it or what they plan to do. I think you wanna kinda look at, like, each release. You know? It's like, we're obviously very pro open source-

Dwarkesh Patel 38:21 Yeah.

Mark Zuckerberg 38:22 ... but I haven't committed that we're gonna, like, release every single thing that we do.

Dwarkesh Patel 38:24 Mm.

Mark Zuckerberg 38:25 But it's basically we, I, like, I'm, I'm just generally very inclined to thinking that open sourcing it is gonna be good for the community and, and also good for us, right? 'Cause we'll, we'll benefit from, from the innovations. Um, but if at, at some point, like, there's some qualitative change in what the, the thing is capable of and we feel like it's just not responsible to open source it, then we won't. But, um, so I don't know. It's, it's, it's all, it's all very difficult to predict.

Dwarkesh Patel 38:51 Yeah. Um, w- what is a kind of qualitative change, like, a specific thing? You're training Llama-5, Llama-4, and you've seen this, and, like, w- w- we're at, you know what? I'm not sure about open sourcing it.

Mark Zuckerberg 39:01 Um, I think that that... It's a little hard to answer that in the abstract because there are negative behaviors that any product can exhibit-

Dwarkesh Patel 39:13 Mm-hmm.

Mark Zuckerberg 39:13 ... that as long as you can mitigate it, it's, like, it's okay. Right? So, um, I mean, there's bad things about social media that we work to mitigate, right? There's bad things about Llama-2 that we spend a lot of time trying to make sure that it's not, like, you know, helping people commit violent acts or things like that, right? I mean, that doesn't mean that it's like a, a, kind of a- a autonomous or intelligent agent. It just means that it's learned a lot about the world and it can answer a set of questions that, um, we think it would be unhelpful for it to answer. Um, so I, um, I don't know. I think the question isn't really what behaviors would it show. It's what things would we not be able to mitigate after it shows that. And, um, and I don't know. I, I, I think that there's so many ways in which something can be good or bad that it's hard to actually enumerate them all up front. If you even look at, like, what we've had to deal with in, in, um, you know, social media and, like, the different types of harms, we've basically gotten to, it's like, there's like 18 or 19 categories of, of harmful things that, that people do, and we've basically built AI systems to try to go identify what those things are that people are doing and try to make sure that that, you know, doesn't happen on our network as much as possible. So, um, yeah. I think you, you can... Over time, I think you'll be able to break down, um, this into more of a taxonomy too, and I, I think that it's, it... This is a thing that we spend time researching too 'cause we wanna make sure that we understand that.

Dwarkesh Patel 40:37 Whoosh. So one of the things I asked Mark is what industrial scale use of LLMs would look like. You see this in previous technological revolutions where, at first, they're thinking in a very small scale way about what's enabled, and I think that's what chatbots might be for LLMs. And I think the large-scale use case might look something like what V7 Go is. And by the way, it's made by V7 Labs, who's sponsoring this episode. So it's like a spreadsheet. You put in raw information, like documents, images, whatever, and they become rows, and the columns are populated by an LLM of your choice. And in fact, I used it to prepare for Mark. So I fed in a bunch of blog posts and papers from Meta's AI research, and as you can see, if, uh, you're on YouTube, it summarizes and extracts exactly the information I want as columns. And obviously, mine is a small use case. But you can imagine, for example, a company like FedEx has to process half a million documents a day. Obviously, a chatbot can't do that. A spreadsheet can because this is just like a fire hose of intelligence in there, right? Anyways, you can learn more about them at v7labs.com/go or the link in the description. Back to Mark. Whoosh. Yeah. Like, uh, it, it seems to me it would be a good idea. I, I would be disappointed in a future where AI systems aren't broadly deployed and everybody doesn't have access to them.

Mark Zuckerberg 41:53 Yeah. Yeah.

Dwarkesh Patel 41:53 Um, at the same time, I wanna better understand the mitigations.

Mark Zuckerberg 41:58 Yeah.

Dwarkesh Patel 41:58 Um, 'cause if the mitigation is the fine-tuning, well, the w- whole thing about open weights is that you can then, um, remove the fine-tuning, which is often superficial on top of these capabilities. Like, if it's like talking on Slack with a biology researcher... And I, again, I think, like, models are very far from this.

Mark Zuckerberg 42:14 Yeah.

Dwarkesh Patel 42:14 They're, right now, they're like Google Search, um, but it's like I can show them my Petri dish, and they can explain, like, "Here's why your, uh, smallpox, uh, sample-"

Mark Zuckerberg 42:20 Yeah.

Dwarkesh Patel 42:21 "... didn't grow. Um, here's what to change." Um, uh, how do you mitigate that? 'Cause th- th- somebody can just, like, fine-tune that in there, right?

Mark Zuckerberg 42:28 Yeah. I mean-That's true. I think a lot of people will basically use the off the shelf model and some people who have basically bad faith are going to try to strip out all the bad stuff, so I do think that's an issue. The, um, the flip side of this is that... And this is one of the reasons why I'm, I'm kind of philosophically so pro open source, is I do think that a concentration-

Dwarkesh Patel 42:55 Yeah.

Mark Zuckerberg 42:55 ... of AI in the future has the potential to be as dangerous as kind of it being widespread. So I think a lot of people are, they, they think about the questions of, "Okay. Well, if it can do this stuff, is it bad for it to be out wild?" Like just in, in-

Dwarkesh Patel 43:11 Yeah.

Mark Zuckerberg 43:11 ... kind of widely available. Um, I think another version of this is like, okay, well, it's probably also pretty bad for one institution to have an AI that is way more powerful than everyone else's AI, right? So if you look at like, like I guess one security analogy that I think of is, um, you know, it doesn't take AI to basically... Okay, there's security holes in so many different things and if you could travel back in time a year or two years, right? It's like that's not AI, it's like you just... Let's say you just have like one year or two years more knowledge of the security holes, you could pretty much hack into like any system, right? So it's not that far-fetched to believe that a, a very intelligent AI would probably be able to identify some holes, um, and, and basically be like a human who could potentially go back in time a year or two and compromise all these systems. Okay, so how have we dealt with that as a society? Well, one big part is open source software that makes it so that when improvements are made to the software, it doesn't just kind of get stuck in one company's products, but it can kind of be broadly deployed to a lot of different systems, whether it's banks or hospitals or government stuff. And like just everyone can kind of... Like as the software gets hardened, which happens because more people can see it and more people can bang on it, um, and there, and there are standards on how this stuff works, um, the world can kind of get upgraded together pretty quickly. And I kinda think that a world where AI is very widely deployed in a, in a way where it's gotten hardened, um, progressively over time, and is one where all the different systems will be in check-

Dwarkesh Patel 44:54 Yeah.

Mark Zuckerberg 44:54 ... in a way that seems like it is fundamentally more healthy to me than one where this is more concentrated. So there are risks on all sides, but I think that that's one risk that I think people, I don't hear them talking about quite as much. I think like there's sort of the risk of like, "Okay, well what if the AI system does something bad?" I, I, I am more like, you know, I stay up at night more worrying, well, what if like some actor that... Whatever. It's like from wherever you sit, there's gonna be some actor who you don't trust. If they're the ones who have like the super strong AI, whether it's some, like other government that we, that, that is sort of like an opponent of, of, of our country or some company that you don't trust or whatev- whatever it is. Um, like I think that that's potentially a much bigger risk.

Dwarkesh Patel 45:43 As in they could like overthrow our government because they have a weapon that like nobody else has?

Mark Zuckerberg 45:50 Could just cause, cause a lot of mayhem. Right? It's, I think it's like... And I, I think the intuition is that this stuff ends up being pretty kind of important and, and, um, and valuable for both kind of economic and, and kind of security and other things. And, um, I don't know. I just think, yeah, if, if like, if someone who you, you don't trust or is an adversary of you gets something that is more powerful, then, um, then I think that that could be an issue. And I think that probably the best way to mitigate that is to have good open source, um, AI that, that basically becomes the standard, um, and in a lot of ways kind of can become the leader. And, um, in that way, it just, it just ensures that it's a much more kind of even and balanced playing field.

Dwarkesh Patel 46:33 Yeah. That seems plausible to me and if that works out, that would be the future I prefer. Um, I guess I want to understand like mechanistically how if somebody was gonna cause mayhem with AI systems, how the fact that there are other open source systems in the world prevents that. Like the specific example of like somebody coming up with a bio weapon. Um, is it just that we'll do a bunch of like R&D in the rest of the world to like figure out vaccines really fast? Like what, what, what's happening?

Mark Zuckerberg 46:57 If you take like the computer, the security one that I was talking about, I think someone with a weaker AI trying to hack into a system that is like protected by a stronger AI will succeed less.

Dwarkesh Patel 47:08 Hmm.

Mark Zuckerberg 47:08 Right? So, so I think that that's, um, I mean, that's like in terms of software security-

Dwarkesh Patel 47:11 How do you know everything in the world is like that? Like what if bio weapons aren't like that?

Mark Zuckerberg 47:16 No, I mean, I don't know that everything in the world is like that. Um, um, I, I think that that's I guess one of the... Bio weapons are one of the areas where I think the people who are most worried about this stuff are focused and, and I think that that's, uh... I think it makes a lot of sense to think about that. Um, the... And I think that there are certain mitigations you can try to not train certain knowledge into the model, right? There's different things, but, um, yeah. I mean, at some level, if you get a sufficiently bad actor and you don't have other AI that can sort of balance them, um, and understand what's going on and what the threats are, then, um, then that could be a risk. So I think that that's one of the things that we need to watch out for.

Dwarkesh Patel 48:03 Hmm. Um, is there something you could see in the deployment of these systems, where you, you observe, like you're training Llama 4 and it's like li- lied to you because it thought you weren't noticing or something. And you're like, "Whoa. What's going on here?" Um, not that you, this is probably not likely with a Llama 4 type system, but is there something you can imagine like that where you'd like be really concerned about deceptiveness and if like billions of copies of things are out in the wild?

Mark Zuckerberg 48:30 Um-Yeah, I mean, I think that that's not necessarily... I mean, right now, it's, we see a lot of hallucinations.

Dwarkesh Patel 48:37 Yeah.

Mark Zuckerberg 48:37 Right? So (laughs) I think it's more, more that. Um, um, I think it's, it's an interesting question, how you would tell the difference between a hallucination and deception. But yeah, I'm not... Look, I mean, I think that there's a lot of risks and things to, to think about. The, um, the flip side of all this is that there are also a lot of... I, I try to in, in, in running our company at least-

Dwarkesh Patel 48:59 Mm-hmm.

Mark Zuckerberg 49:00 ... balance what I think of as these longer term theoretical risks, um, with what I actually think are quite real risks that exist today. So like when, when you talk about deception, the form of that that I worry about most is people using this to generate misinformation and then like pump that through whether it's our networks or others. So the way that we've basically combated a lot of this type of harmful content is by building AI systems that are smarter than the adversarial ones. And I guess this is part of, this kind of informs part of my theory on this, right? Is if you look at like the different types of harm that people do or try to do through, through social networks, um, there are ones that are not very adversarial. So for example, like, uh, hate speech I would say is not super adversarial in the sense that like people aren't getting better at being racist, right? They're just like, it's y- y- you just like, okay, if you, you kind of... That's one where I think the AIs are generally just getting way more sophisticated faster than people are at those issues. So we have, and we have issues both ways. It's like people do bad things that whether they're trying to incite violence or something. Um, but we also have a lot of false positives, right? So where we, where we basically censor stuff that we shouldn't, and-

Dwarkesh Patel 50:22 Mm-hmm.

Mark Zuckerberg 50:22 ... I think understandably make a lot of people annoyed. So I think having an AI that just gets increasingly precise on that, that's gonna be good over time. But let me give you another example, which is like nation-states trying to interfere in elections. That's an example where they're absolutely, they have cutting edge technology and absolutely get better each year. So we block some technique, they learn what we did, they come at us with a different technique, right? It's not like a person trying to, you know, I don't know, say, say mean things, right? It's like, it's, it's, they, they're, they're basically, they have a goal, they're sophisticated, they have a lot of technology. Um, in those cases, I, I still think the ability to kind of have our AI systems grow and in sophistication at a faster rate than theirs have, it's an arms race, but I think we're at least currently winning that arms race. Um, so I don't know. I think that that's... But this is like a lot of the stuff that I, that I spend time thinking about is like, okay, yes, it is possible that whether it's Llama4 or Llama5 or Llama6, yeah, we need to think about like what behaviors we're, we're observing. And it's not just us. I think part of the reason why you make this open source is that there are a lot of other people who study this too. So yeah, we wanna see what other people are observing, what we're observing, what we can mitigate, and then we'll make our, our assessment on whether we can make it, um, open source. But I, I think for the foreseeable future, I'm, I'm optimistic we will be able to, and in the near term, I don't wanna take our eye off the ball of what are actual bad things that people are trying to use the models for today, even if they're not existential, but they're like, they're like pretty bad kind of day-to-day harms that we're familiar with in running our services. Um, that's actually a lot of what we have to, I think, spend our time on as well.

Dwarkesh Patel 52:07 Yeah, yeah. Um, actually, I, I, I found the synthetic data thing really curious. Uh, I'm, I'm actually interested in why you don't think... Uh, like current models, it makes sense why there might be an asymptote with just doing the synthetic data again and again.

Mark Zuckerberg 52:20 Yeah.

Dwarkesh Patel 52:20 If they get smarter and you use this kind of techniques you talk about in the paper or the blog post that's coming out, um, o- on the day this will be released where it, it goes to the thought chain that is the most, um, correct, why you, why this wouldn't like lead to a loop that over, of course it wouldn't be overnight, but over many months or years of training, potentially with a smarter model, it gets smarter, makes better output, gets smarter and so forth?

Mark Zuckerberg 52:44 Um, well, I think it could within the parameter of whatever the model architecture is. It's just that like at some level, I don't know, I, I think like today's eight billion parameter models, I just don't think you're gonna be able to get to be as good as the state-of-the-art multi-hundred billion parameter models that are incorporating new research into the architecture itself. Um...

Dwarkesh Patel 53:12 But those will be open source as well, right?

Mark Zuckerberg 53:14 Well, yeah, but I think that that's, i- if- if... I mean, subject to all of the-

Dwarkesh Patel 53:19 Yeah.

Mark Zuckerberg 53:20 ... questions that we just talked about.

Dwarkesh Patel 53:21 Sure, sure.

Mark Zuckerberg 53:21 But yes, I mean, we would, we would hope that that'll be the case, but I, but I think that at each point, I don't know, it's like when you're building software, there's like a ton of stuff that you can do with software, but then at some level you're constrained by the chips that it's running on, right? So there are always gonna be different physical constraints, and it's like how big are the models is gonna be constrained by how much energy you can get and, um, and use for inference. Um, so I guess I'm simultaneously very optimistic that this stuff will continue to improve quickly and also a little more measured than I think some people are about kind of it's... I, I, I just don't think the runaway case is like a particularly likely one.

Dwarkesh Patel 54:15 Hmm. I, I think it makes sense to keep your options open. Like there's so much we don't know.

Mark Zuckerberg 54:20 Mm-hmm.

Dwarkesh Patel 54:20 Um, there's a case in which like it's really important to keep the balance of power so nobody becomes like a-

Mark Zuckerberg 54:24 Yeah.

Dwarkesh Patel 54:24 ... totalitarian dictator. There's a case in which like you don't wanna open source, uh, the architecture because like China's catch, can use it to catch up to A- America's AIs and like there is an intelligence explosion and they like win that. Um, yeah. A lot of things seem possible, just like keeping your options open, considering all of them, um,

Dwarkesh Patel 54:40 seems reasonable.

Mark Zuckerberg 54:40 Yeah.

Dwarkesh Patel 54:41 Um-Uh, let's talk about some other things. Uh-

Mark Zuckerberg 54:43 Go for it.

Dwarkesh Patel 54:44 ... okay. Uh, metaverse, what time period in human history would you be most interested in going into? Ah, 100,000 BCE to now, you just wanna see what it was like.

Mark Zuckerberg 54:53 Well, as to the past?

Dwarkesh Patel 54:54 Huh?

Mark Zuckerberg 54:55 It has to be the past?

Dwarkesh Patel 54:55 Oh, yeah, it has to be the past.

Mark Zuckerberg 54:57 Um, I don't know. I mean, I, I have the periods of time that I'm interested. I mean, I'm really interested in American history and classical history and, um, I'm really interested in the history of science too. So I actually think seeing and trying to understand more about how some of the big advances came about. I mean, all we have are, like, somewhat limited writings about some of that stuff. I'm not sure the metaverse is gonna let you do that 'cause I mean, it's, um... You know, we can't... Mo- It's gonna be hard to, to kinda go back in time for things that we don't have records of. But, uh, I'm actually not sure that going back in time is gonna be that e- that, that important of a thing for the... I mean, I think that it's gonna be cool for, like, history classes and stuff. But, um, that's probably not the use case that I'm most excited about for the, for the metaverse overall. I mean, it's, um... I think the main thing is just the ability to feel present with people no matter where you are. I think that that's gonna be killer. I mean, there's, um... I mean, in the AI conversation that we, that we're having, I mean, it's, uh... You know, so much of it is about physical constraints that kind of underlie all, all of this, right? And you wanna move... I think one lesson of technology is you wanna move things from the physical constraint realm into software as much as possible because software is so much easier to build and, and evolve. And, like, you can democratize it more because, like, not everyone is gonna have a data center. But, like, uh, a lot of people can, can kind of write code and take open source code and, and modify it. Um, the metaverse version of this is, I think, enabling realistic digital presence is going to be just an absolutely huge difference for, um, for making it so that, um, people don't feel like they have to physically be together for as many things. Um, now, I mean, I think that there are gonna be things that are better about being physically together. Um, so it's not... I mean, these things aren't binary. It's not gonna be like, "Okay, now it's, you don't need to do that anymore." But, um, but overall, I mean, I, I think that this, it's just gonna be really powerful for, for socializing, for feeling connected with people, for working, um, for, I don't know, parts of industry, for medicine, for, like, a l- a l- like, so many things.

Dwarkesh Patel 57:20 I, I wanna go back to something you said at the beginning of the conversation where, um, you didn't sell the company for a billion dollars.

Mark Zuckerberg 57:25 Uh-huh.

Dwarkesh Patel 57:25 And like with the metaverse, you knew we were gonna do this even though the, uh, the, the market was hammering you for it. And then I'm actually curious, like, what is the source of that edge? And you said, like, "Oh, values. I have this intuition." But, like, everybody says that, right? Like, what do... If you had to say something that's specific to you, what is... How would you express what that is? Like, why, wh- why were you so convinced about the metaverse?

Mark Zuckerberg 57:46 Um... Uh, well, I think that those are different questions. So what, I mean, what, what are the things that, that kinda power me? Um, I think we've talked about a bunch of the themes. So it's... I mean, I, I just really like building things. Um, I specifically like building things around how people communicate and sort of understanding how people express themselves and how people work, right? It's... When I was in college, I was, I was... I studied computer science and psychology. I think a lot of other people in the industry studied, studied computer science, right? So, um, it's, uh, it's always been sort of the intersection of those two things for me. But I think it's also sort of this, like, really deep drive. I don't know how to explain it, but I just feel like in... like constitutionally, like, I'm doing something wrong if I'm not building something new. Right? And, and, um, so I think that there's, like... You know, even when we're putting together the business case for, you know, investing, like, $100 billion in AI or some huge amount in the metaverse. It's like, yeah, I mean, we have plans that I think make it pretty clear that if our stuff works, it'll be a good investment. But, like, you can't know for certain from the outset. And, um, so there's all these arguments that people have, you know, whether it's like, you know, with advisors or, or different folks. It's like, "Well, how, how could you..." Like, it's... How, how are you confident enough to do this? And it's like, "Well, the day I stop trying to build new things, I'm just done. I'm gonna go build new things somewhere else," right? It's like, um, it's like it is... I, I'm fundamentally incapable of running something or in my own life and, like, not trying to build new things that I think are, are interesting. It's like that's not even a question for me, right? It's not like whether, like, whether we're gonna go take a swing at, like, building the next thing. It's like, it's... I, like, I'm, I'm just incapable of not doing that. Um, and I don't know. I, and I'm kinda like this in, like, all the different aspects of my life, right? It's like we built this, like... Our family built this ranch in Kauai and, like, I just, like, worked to, like, design all these buildings and, like, kinda try to... Or, like, we started raising cattle and I'm like, "All right, well, I wanna make, like, the best cattle in the world," right? So it's like how do we... Like, how do we architect this so that way we can figure this out and, like, and build, like, all the stuff up that we need to to try to do that. Um, so I don't know. Uh, th- that's me. Um, what was the other part of the question?

Dwarkesh Patel 1:00:32 Look, Meta is just a really amazing tech company, right? They have all these great software engineers and even they work with Stripe to handle payments, and I think that's just a really notable fact that Stripe's ability to engineer these checkout experiences is so good that big companies like Ford, Zoom, Meta, even OpenAI, they work with Stripe to handle payments. Because just think about how many different possibilities you have to handle. If you're in a different country, you'll pay a different way, and if you're buying a certain kind of item, that might affect how you decide to pay. And Stripe is able to test these fine-grained optimizations across tens of billions of transactions a day to figure out what will convert people. And obviously, conversion means more revenue for you.And look, I'm not a big company like Meta or anything, but I've been using Stripe since long before they were advertisers. Stripe Atlas was just the easiest way for me to set up an LLC, and they have these payments and invoicing features that make it super convenient for me to get money from advertisers. And obviously, without that, it would have been much harder for me to earn money from the podcast. And so it's been great for me. Go to stripe.com to learn more. Thanks to them for sponsoring the episode. Now, back to Mark. I'm not sure about that. I- I'm actually curious about something else, which is, um, um, so a 19-year-old Mark, um, reads a bunch of like antiquity and classics, uh, high school, college. What important lesson did you learn from it? Not just interesting things you found, but like, there aren't that many tokens you consume by the time you're 19. A bunch of them were about the classics. Clearly, that was important in some way.

Mark Zuckerberg 1:01:57 Not that many tokens you consume. (laughs) Um, I don't know. That's a good question. I mean, one of the things that I thought was really fascinating is, um... So when Augustus was first... so he, he became emperor, and, um, and he was trying to establish peace, and the- there was no real conception of peace at the time. Like, the people's- people's understanding of peace was, it is the temporary time between when your enemies will inevitably attack you again, so you get like a short rest. And, and he had this view, which is like, "Look, like, we want to change the economy from instead of being so mercenary and like, and kind of militaristic, to like actually this positive sum thing." It's like a very novel idea at the time. Um... I don't know. I think that- that there's like something that's just really fundamental about that. It's like in terms of the, the bounds on like what people can conceive at the time of like what are rational ways to work. And, um... I don't know. Going back to like... and this applies to both the metaverse and the AI stuff, but like, a lot of investors and just different people just can't wrap their head around why we would open source this. And it's like, are y- like, like... I don't understand. It's like open source? That must just be like the temporary time between which you're making things proprietary, right?

Dwarkesh Patel 1:03:35 (laughs)

Mark Zuckerberg 1:03:35 And it's, um... But, but I actually think it's like this very profound thing in tech that has actually... it- it creates a lot of winners, right? And it's... and, and, um, so... I don't know. I don't want to strain the analogy too much, but- but I do think that there's, um... there's a lot of times I think ways where you can... that are just like models for building things that people can't even... like they just like often can't wrap their head around how that would be a valuable thing for people to go do, or like a reasonable state of the world that it's... I m- I mean, it's, uh... I think that there's more reasonable things than people think.

Dwarkesh Patel 1:04:20 That's super fascinating. Um, can I give you my answer, what I was thinking-

Mark Zuckerberg 1:04:24 Sure.

Dwarkesh Patel 1:04:24 ... what you might have gotten from it? Um, I to- this is probably totally off, but, um, just how young some of these people are who have very important roles in the empire. Like Caesar Augustus, like by the time he's 19, he's actually incredibly- one of the most prominent people in Roman politics, and he's like leading battles and forming the Second Triumvirate. I wonder if you're like, the 19-year-old is like, "I can actually do this because like Caesar Augustus did this."

Mark Zuckerberg 1:04:45 I- I think that that's... I think that's an interesting example both from a lot of history and American history too.

Dwarkesh Patel 1:04:50 Yeah.

Mark Zuckerberg 1:04:50 I mean, it's, um... I mean, one of my favorite quotes i- it's this Picasso quote that, "All children are artists, and the challenge is how do you remain an artist when you grow up?" And it's like basically I think 'cause when you're younger, I think you- it's just easier to have kind of wild ideas and you're not... you know, you have no... there are all these analogies to the innovator's dilemma that exist in your life as well as your company or whatever you've built, right? So, you know, you're kind of earlier on in your trajectory. It's easier to pivot and- and take in new ideas without it disrupting other commitments that you've made to- to different things. And, um... so I don't know. I think that that's an interesting part of- of running a company is like how do you... how do

Mark Zuckerberg 1:05:36 you kind of stay dynamic?

Dwarkesh Patel 1:05:37 Hmm. Um, going back to the investors in open source, uh, the $10 billion model. Suppose it's- it's totally safe, you've done these evaluations, and um, unlike in this case, the evaluators can also fine-tune the model, um, which hopefully will be the case in future models. Uh, ah, would you open source that, the $10 billion model?

Mark Zuckerberg 1:05:55 Well, I mean, as long as it's helping us, then yeah.

Dwarkesh Patel 1:05:57 But w- would it? Like the $10 billion of R&D and then now it's like open source for anybody.

Mark Zuckerberg 1:06:00 Well- well, I think h- here's I think a question which w- we'll- we'll have to evaluate this as- as time goes on too, but, um... we have a long history of open sourcing software, right? We don't tend to open source our product, right? So it's not like we ta- we don't take like the code for Instagram and make it open source, but we take like a lot of the low-level infrastructure and we make that open source, right? The- the probably the biggest one in our history was Open Compute Project where we took the designs for kind of all of our, um, our servers and network switches and data centers and made it open source and ended up being super helpful because, you know, I mean, a lot of people can design servers but now like the industry standardized on our design, which meant that the supply chains basically all got built out around our design and the volumes went up so it got cheaper for everyone and saved us billions of dollars. So, awesome, right? Okay, so there's multiple ways where open source I think could be helpful for us. One is if people figure out how to run the models more cheaply. Well, if we're gonna be spending tens or like a hundred billion dollars or more over time, um, on all this stuff, so if we can do that 10% more effectively we're saving billions or tens of billions of dollars. Okay, that's probably worth a lot by itself. Um-... especially if there's other competitive models out there. It's not like our thing is- is, like, be giving away some kind of crazy advantage. Um...

Dwarkesh Patel 1:07:22 So, is it w- your view that l- the training will be commodified?

Mark Zuckerberg 1:07:28 (sighs) Um, I think there's a bunch of ways that this could play out. That's one. The, um, the other is, is that, so commodity kind of implies that it's gonna get very cheap because, um, 'cause there's lots of options. The other direction that this could go in is qualitative improvements. So, um, so you mentioned fine-tuning, right? It's like right now, it's, it's, um, you know, it's pretty limited what you can do with fine-tuning major other models out there and there are some options, but generally not for the biggest models. Um, so I think being able to do that and- and be able to kind of do different app-specific things or use case-specific things or build them into specific tool chains, um, I think will not only enable kind of more efficient development. It could enable qualitatively different things. Um, here's one analogy on this is, um... So, one thing that I think generally sucks about the mobile ecosystem is that, like, you have these two gatekeeper companies, Apple and Google, that can tell you what you're allowed to build. And there are lots of times in our history... So there's the economic version of that, which is like, "All right, we built something," and they're just like, "I'm gonna take a bunch of your money." But then there's the, there's the, um, the qualitative version, which is actually what kind of upsets me more, which is there's a bunch of times when we've launched or wanted to launch features and then Apple's just like, "Nope, you're not launching that." I'm just like, "That sucks."

Dwarkesh Patel 1:08:55 (laughs)

Mark Zuckerberg 1:08:55 Right? And, um, so the question is, what is it, like, are we kind of set up for a world like that with AI where, like, you're gonna get a handful of companies that run these closed models that are gonna be in control of the APIs and therefore are gonna be able to tell you what you can build? Um, well, for one, I can say for- for us, it is worth it to go build a model ourselves to make sure that we're not in that position.

Dwarkesh Patel 1:09:20 Mm-hmm.

Mark Zuckerberg 1:09:20 Right? Like, I don't want any of those other companies telling us what we can build. Um, but from an open source perspective, I think a lot of developers don't want those companies telling them what they can build either. Um, so the question is, what is the ecosystem that gets built out around that? What are interesting new things? How much does that improve our products? Um, I think there's a lot of cases where if this ends up being, like, you know, like our databases or caching systems or architecture, we'll get valuable contributions from the community that will make our stuff better, and then our app-specific work that we do will still be so differentiated that it won't really matter, right? It's like we'll- we'll be able to do what we do, we'll benefit all the systems, ours and the community's will be better because it's open source. There is one world where, um, maybe it's not that. I mean, maybe the model just ends up being more of the product itself. In that case, then I think it's, um, it's a trickier economic calculation around whether you open source that because then you- you are kind of commoditizing yourself a lot. But I don't, uh... From what I can see so far, it doesn't seem like we're in that zone.

Dwarkesh Patel 1:10:26 Um, would you expect to earn significant revenue from licensing your model to the cloud providers so they have to pay you a fee to actually serve the model?

Mark Zuckerberg 1:10:33 Um, we- we wanna have an arrangement like that, but I don't know how significant it'll be. And we have this, um, this is basically our license for- for LLaMA.

Dwarkesh Patel 1:10:44 Yeah.

Mark Zuckerberg 1:10:45 Um, you know, in- in a lot of ways, it's- it's like a very permissive open source license, except that we have a limit for the largest companies using it. And this is why we put that limit in, is w- we're not trying to prevent them from using it. Uh, we just want them to come talk to us because if they're gonna just basically take what we built and resell it and make money off of it, then it's like, okay, well, if- if you're like, you know, Microsoft Azure or Amazon, then yeah, if you're gonna be re- reselling the model, then we should have some revenue share on that so just come talk to us before you go do that. And that's how that's played out. So for LLaMA 2 it's, um, I mean, we basically just have deals with all these major cloud companies and LLaMA 2 is available as a hosted service on all those clouds, and, um, I assume that as we, as we release bigger and bigger models, that'll become a bigger thing. It's not the main thing that we're doing, but I- I just think it further assert. If those companies are gonna be selling our models, it makes sense that we should, you know-

Dwarkesh Patel 1:11:38 Yeah.

Mark Zuckerberg 1:11:38 ... share the upside of that somehow.

Dwarkesh Patel 1:11:39 Yeah. Um, with- with the rest of the other open source dangers, I- I think you have a- a, like, genuine legitimate points about the balance of power stuff, um, and potentially, like, the harms you can get rid of because we have better alignment techniques or something. Um, I wish there were some sort of framework that Meta had, like other labs have this where they say like, "If we see this as a concrete thing, then the- the- that's a no go on the open source or, like, w- even potentially on deployment." Um, just like writing it down so, like, uh, the company is ready for it.

Mark Zuckerberg 1:12:06 Yeah.

Dwarkesh Patel 1:12:06 Uh, people have expectations around it and so forth.

Mark Zuckerberg 1:12:09 Yeah. No, I think that that's a fair point on the existential risk side.

Dwarkesh Patel 1:12:12 Yeah.

Mark Zuckerberg 1:12:12 Right now, we focus more on the types of risks that we see today, which are more of these content risks. So, you know, we have lines on we don't want the model to be basically doing things that are helping people commit violence or fraud or, you know, just harming people in different ways. So, um, in practice for today's models, and I would guess the next generation and maybe even the generation after that, I- I think while it is somewhat more maybe intellectually interesting to talk about the inte- the existential risks, I- I actually think the- the real harms that need more energy being mitigated are things that are going to, like, to h- have someone take a model and do something to hurt a person with today's parameters of- of- of... and kind of the types of kind of more mundane harms that we see today, like people kind of committing fraud against each other or things like that. So, um, that, I- I just don't want to shortchange that. I think we- we have a responsibility to make sure we do a good job on that.

Dwarkesh Patel 1:13:17 Yeah. Meta's a big company, it can handle both.

Mark Zuckerberg 1:13:18 Yeah.

Dwarkesh Patel 1:13:19 Um, uh, o- okay, so a- a- as far as the open source goes, I'm actually curious if you think the impact of the open source from Pytorch, React, Open Compute, these things, has been bigger for the world than even the social media aspects of Meta. 'Cause I've, like, talked to people who use these services who think, like, it's plausible. 'Cause a- a big part of the internet runs on these things.

Mark Zuckerberg 1:13:37 Um, ah, it's an- it's an interesting question. I mean, I think a- almost half the world uses our-

Dwarkesh Patel 1:13:43 Yeah. That's the